High-dimensional Data Stream Control Chart And Application Based On Bayesian Method

Due to the diversification of business development and the expansion of production scale in today's business,the data generated during the production process is not only large in terms of observation and measurement,but also has higher and higher data dimensions.Metering process control can effectively use the out-of-control quality characteristics to monitor possible abnormal conditions.However,in the case of such high-dimensional data flow,the traditional production quality process control maps are generally effective.When an abnormality occurs,alarm signals cannot be issued quickly.Therefore,in order to improve the efficiency of production process control,it is necessary to have an existing control chart method.Make improvements.In this paper,the Bayesian variable selection method is combined with multivariate control charts to detect changes in the mean value of the sparsity of the affected stream in the high-dimensional process with a normal distribution.Based on the control chart method of high dimensional Bayesian variable selection,the data is firstly screened by the high dimensional Bayesian variable selection method to realize the data dimensionality reduction process,and then the control limits and process control are calculated according to the variable selected data set.Statistics draw control charts to achieve the control process.Through simulation studies,the monitoring efficiency of the process control statistics before and after Bayesian variable selection was compared and analyzed to study the improvement effect of the high-dimensional Bayesian variable selection method on process control statistics.In a further case study,the simulation results were analyzed empirically using data sets collected from sensors during modern semiconductor manufacturing.Studies have shown that when there is an average run length?ARL?with anomalous data after a fixed steady-state average run length?ARL₀?,when pa?5,the monitoring effect of the control chart before and after variable screening has little difference;when pa??5,40?,the ARL based on the Bayesian variable selection control chart method is much smaller than before the variable selection,and the Bayesian variable selection method has a very significant effect on the T_newew process control;when pa?40,The Bayesian variable selection method gradually reduces the process control effect of T_new.In general,the high-dimensional Bayesian variable selection method has a better effect on process control,and can more quickly issue an alarm when a fault occurs,thereby improving the efficiency of the actual production process.